Mentioned in "TextCat and Confidence" as a possible confidence measure, the ratio of first to second result scores was not investigated as part of the overall TextCat improvement.

The idea is that if the best score is x, then it is more likely to be correct if the second best score is x+50% than if the second best score is x+10%. The new config will require a gap of 6% (as less than that is too ambiguous to even consider).

This should be a quick test; if it turns out to be worthwhile, implementing it properly will be more work.